mpii-crops-512
收藏资源简介:
MPII单个人像裁剪(512×512)数据集,适用于Instruct Pix2Pix准备的数据集。每个样本包含一对图片和一个编辑提示字符串。输入图片是经过填充和调整大小的512×512像素的裁剪图,没有点标记。输出图片是相同的裁剪图,但包含了可见关键点的标记点。此外,还包括图片ID、人员ID、裁剪图片的宽度和高度以及文件扩展名等额外字段。图片裁剪是基于关键点计算得到的边界框,并填充至正方形后调整大小。标记点是在调整大小后添加的固定半径的实心圆。
MPII single-person cropped (512×512) dataset, prepared for use with Instruct Pix2Pix. Each sample consists of an image pair and an editing prompt string. The input images are cropped, padded and resized to 512×512 pixels, with no point markers included. The output images are identical to the cropped input images, but with visible key point markers overlaid. Additionally, the dataset provides supplementary metadata fields including image ID, person ID, width and height of the cropped image, and file extension. The cropping process is based on the bounding box calculated from the key points, followed by padding to form a square dimension and resizing to 512×512 pixels. The point markers are solid circles with a fixed radius, which are added after the resizing step.
MPII Single-Person Crops (512×512) — Instruct Pix2Pix Ready 数据集概述
数据集基本信息
- 数据集名称:MPII Single-Person Crops (512×512) — Instruct Pix2Pix Ready
- 存储位置:https://huggingface.co/datasets/lukalafaye/mpii-crops-512
- 下载大小:4,244,778,782 字节
- 数据集大小:759,995,781 字节
数据特征
核心特征
- edit_prompt:指令字符串,用于Instruct Pix2Pix流程
- input_image:输入图像,512×512填充裁剪(无标记点)
- output_image:输出图像,相同裁剪带有关键点标记(仅显示可见关键点)
附加字段
- image_id:原始图像词干(来自
_rXX前的文件名) - person_id:来自
_rXX的人物索引 - width/height:输入裁剪图像的尺寸
- ext:文件扩展名
数据划分
- 训练集:17,265个样本,682,695,445字节
- 验证集:1,919个样本,77,300,336字节
数据处理说明
- 裁剪基于关键点计算的边界框居中
- 填充为正方形后调整至512×512分辨率
- 标记点为恒定半径的实心圆,在调整大小后添加




